Multi-Disciplinary Design Optimisation

This presentation was made by Professor Andy Keane of the University of Southampton at the October 2017 Optimisation Working Group Special Interest Group meeting. NAFEMS members are invited to get involved and participate in these events. More information on the Optimisation Special Interest Group can be found here.

Design improvement generally involves two related activities: deciding what things to change (and by how much) and predicting the effects of such changes (typically by experiment, practical or computational). In this talk we focus on the first of these issues assuming the designer has access to suitable means of evaluating any proposed new design. This is the traditional role of optimizers in engineering design, and while there are many powerful optimization tools readily available (we often use surrogate based methods) their use requires two important precursors: first one must be able to represent desired changes as vectors of numbers and secondly one must be able to use such vectors to automatically generate sufficient information for predictive methods to be applied. This is often more difficult than might be supposed, particularly if it is the shape of an artefact that is to be optimized – interesting geometry generation and meshing is sometimes far from simple. This talk relates some of our experiences with surrogate based search tools over geometrically complicated problems.